Conversational Agents and Knowledge - FAQs

Building a conversational Agent

How do I build a conversational Agent?

Create an Agent workflow, add the trigger for its conversational channel, and add a Conversational Agent activity. Configure the model, instructions, memory, and any tools, then test before publishing. See building a conversational Agent.

Which conversational channels can I use?

The building guide describes web chat, WhatsApp message handling, and incoming telephone calls. Select the corresponding trigger and configure its required connection and channel settings. See conversational triggers.

What is the difference between persona settings and Agent instructions?

Persona and trigger settings define the conversational identity and welcome experience. Instructions in the Conversational Agent activity define responsibilities, business rules, response behavior, and tool-use guidance. See AI persona and Agent behavior.

How do I configure an AI model?

Configure the model in the Conversational Agent activity using a customer-managed provider connection or Integration Gateway services. The available options depend on the provider and model. See model configuration and Integration Gateway.

How does conversation memory work?

Memory preserves context across interactions. Configure history retention and summarization to balance continuity with token usage. See conversation memory.

Can a conversational Agent execute business actions?

Yes. Connect tools that retrieve records, invoke workflows, call integrations, or perform other configured actions. Give each tool clear inputs and a defined purpose, and test its success and failure paths. See tools and actions.

How do I test and publish the Agent?

Use Run in Agent Builder to test the configured experience. Validate conversations, knowledge retrieval, tools, and voice where enabled, then publish. Testing and production URLs are available through the trigger configuration where supported. See testing and publishing.

Where can I review conversations and execution errors?

Use the Agent's Chat History for messages and responses. Use Reports → Automation for workflow execution, tool calls, and errors; Debug Runs can provide additional detail. See execution and conversation history.

Enterprise knowledge and ingestion

How do I make an Agent answer using our documents and websites?

Ingest the sources for the owning Agent, then connect a Knowledge Base activity to the Conversational Agent's Tools connector. Configure its Query for AI-generated input so retrieval follows the user's request. Ingestion alone does not enable conversational retrieval. See knowledge grounding and using enterprise knowledge.

How do I ingest files?

Configure a File Upload ingestion source, upload the content, review the processing configuration, and start ingestion. Check that the required files were successfully processed. See file upload ingestion.

How do I ingest website content?

Configure a Website ingestion source, supply its source settings, and review the content selected for processing before starting. Monitor processed URLs to identify skipped or failed content. See website ingestion.

What infrastructure does knowledge ingestion require?

The ingestion service uses the additional vector and graph data services listed in the server requirements. Provision these alongside the platform's other required services. See ingestion architecture and additional data services.

How do I monitor or rerun an ingestion?

Open the Ingestions dashboard, expand the source, and inspect the run's status and processed content. Use the rerun action to create a new run when content changes or processing needs retrying. See ingestion operations.

How do I remove outdated knowledge?

Use the ingestion operations to remove unwanted processed content and manage obsolete runs. Review what is being removed and verify the Agent's responses afterward. See deleting processed content and knowledge maintenance.

Why is the Agent not using the expected knowledge?

Check that the source belongs to the intended Agent, ingestion completed, and the required content was processed. Verify the Knowledge Base tool connection, AI-generated Query configuration, and instructions for using knowledge. Test a representative question and inspect execution history. See knowledge grounding, query configuration, and ingestion operations.

Voice and conversation behavior

How do I enable voice conversations?

Enable voice in the channel's conversational settings and configure the required speech services. Browser voice also uses real-time communication infrastructure. See voice capabilities; EKS deployments have browser voice and TURN configuration.

How do I adjust interruptions, listening, or response delays?

Use the advanced voice settings for speech detection, interruption thresholds, noise filtering, and response delay. Start with defaults and adjust settings for a specific observed problem. See advanced voice settings.

How do I configure idle prompts and conversation timeouts?

Configure Advanced → Inactivity & Session Settings in the Conversational Agent activity. Check the documented support for the selected channel and test the timing behavior. See inactivity and session settings.

How can I improve response quality?

Use focused instructions, trusted knowledge, and only the tools needed for the task. Test typical requests, missing information, tool failures, and retrieval failures; review conversation and execution history to guide adjustments. See conversational Agent best practices.